[pull] main from NVIDIA:main#604

Merged

pull[bot] merged 7 commits into

phu0ngng:mainfrom

May 12, 2026

Commits on May 12, 2026

Disable the RHT fusion for non-SM100 family devices (#2968)

ptrendx
and
pre-commit-ci[bot]
authored
[PyTorch] Expose function to bulk-allocate tensors backed by the same buffer (#2900)

authored
fix(CP, FA): the conditional logic in the FA version contains a vulnerability when processing the output of Flash Attn forward pass (#2825)
zhujian19891203
authored
[Common] Use specialized unfused MXFP8 cast kernels by default (#2958)

authored
Build Docs fix (#2982)
vthumbe1503
authored
[JAX] Add wait per multi-proc cleanup in `L0_jax_distributed_unittest` (#2979)
phu0ngng
authored
Avoid CPU offload wait_event for validation (#2793)

vasunvidia
and
pre-commit-ci[bot]
authored